Human-Led AI Services: the lessons learned from composable enterprise and API-led Connectivity

Introduction

In the swiftly evolving landscape of technology, the convergence of two powerful paradigms, API-Led Connectivity and AI as a Service (AIaaS), has emerged as a beacon of innovation. This fusion of strategies redefines how we integrate artificial intelligence (AI) capabilities and enhances the scalability, efficiency, and accessibility of these services.

API-Led Connectivity revolutionizes the integration of diverse systems and applications by utilizing APIs as the building blocks for connecting and exposing data and functionalities. It embraces a structured approach with layers dedicated to system/API, process, and experience, thereby promoting reusability, scalability, and efficient collaboration.

On the other side, AI as a Service (AIaaS) offers the delivery of AI capabilities as cloud-based services, allowing businesses and developers to access and implement AI functionalities without the need for extensive infrastructure and expertise. This paradigm shift democratizes AI and offers benefits like cost-efficiency, scalability, and rapid integration.

In this article, we are proposing an approach for designing AI Services APIs using parallelism with an API-led connectivity paradigm. Let’s explore it!

What is API-Led Connectivity?

In the ever-evolving digital landscape, enterprises confront the persistent challenge of integrating diverse systems and applications to optimize operations and deliver seamless user experiences. To tackle the intricacies of integration effectively, a prevailing methodology known as API-led connectivity has emerged, centering on the utilization of Application Programming Interfaces (APIs) to facilitate communication and data exchange among various components. Instead of constructing large, monolithic systems, API-led connectivity deconstructs integration into manageable, discrete entities in the form of APIs that represent specific business functionalities.

The API-led approach encompasses three distinct layers:

• System/API Layer: Positioned at the foundation, this layer concentrates on individual systems and applications, where APIs are meticulously fashioned to expose their specific functionalities. Each API is dedicated to a well-defined service or capability, thereby fostering independent accessibility.
• Process Layer: Intervening between the foundational and topmost layers, the process layer orchestrates and amalgamates functionalities from multiple APIs in the system layer. This strategic interconnection empowers the creation of composite services and business processes, facilitated by the seamless connection and reuse of existing APIs.
• Experience Layer: Occupying the zenith, the experience layer directly interfaces with end-users or external systems. It encompasses APIs tailored to provide bespoke experiences, such as mobile applications, web portals, or other user interfaces.

?The myriad benefits of reusability, flexibility, and improved collaboration make API-led connectivity invaluable for enterprises seeking to establish a responsive and agile IT infrastructure. However, addressing the challenges of governance and external API dependencies is imperative to fully harness the potential of API-led connectivity. By doing so, enterprises can unlock the true power of seamless integration, culminating in the delivery of exceptional user experiences in the digital era.

And, what about AIaaS?

AIaaS refers to the delivery of artificial intelligence capabilities and functionalities as a cloud-based service over the Internet. This approach enables businesses and developers to access and utilize AI-powered tools, algorithms, and models without requiring extensive in-house infrastructure and expertise.

AIaaS providers typically offer a range of AI-related services, such as natural language processing, image and video recognition, machine learning, speech recognition, sentiment analysis, and more. These services are made accessible through well-defined APIs, allowing developers to integrate AI functionalities directly into their applications or services.

The advantages of AI as a Service include:

• Cost-Efficiency: By utilizing AIaaS, organizations can avoid the significant upfront costs associated with building and maintaining their AI infrastructure. Instead, they pay for the AI services they consume on a pay-as-you-go or subscription-based model, making it cost-effective for businesses of all sizes.
• Scalability: AIaaS platforms are designed to handle varying workloads, providing the flexibility to scale resources up or down based on demand. This ensures that applications leveraging AI capabilities can adapt to changing user needs and data volumes.
• Speed and Time-to-Market: Implementing AI functionalities can be complex and time-consuming. AI as a Service simplifies the integration process, allowing developers to incorporate AI into their applications and reduce time-to-market quickly.
• Access to Advanced AI Capabilities: AIaaS providers often offer state-of-the-art AI models and algorithms, which might be difficult for individual organizations to develop or access on their own. This allows businesses to leverage cutting-edge AI technologies without investing in extensive research and development.
• Easy Integration: AIaaS platforms typically offer well-documented APIs, making it straightforward for developers to integrate AI functionalities into their existing applications or services, regardless of the programming language or framework used.

Despite the benefits, there are also some considerations and drawbacks to using AI as a Service:

• Data Privacy and Security: Depending on the sensitivity of the data being processed by AIaaS platforms, there may be concerns about data privacy and security. Businesses should carefully evaluate the data handling policies and security measures of the AIaaS provider.
• Performance and Latency: The performance and response times of AIaaS solutions are subject to internet connectivity and data transmission speeds. In some cases, real-time applications may face challenges due to network latency.
• Regulatory Compliance: Organizations operating in highly regulated industries must ensure that using AIaaS complies with industry-specific regulations and data governance policies.

The Confluence of Strategies: Human-Led AI Services

By combining these two paradigms, businesses can tap into AI services without the need for extensive in-house infrastructure or expertise. The flexibility and modularity offered by AIaaS APIs empower developers to choose specific AI services that suit their requirements, optimizing cost-effectiveness and promoting reusability. Moreover, AIaaS APIs expedite AI adoption, reducing development complexities, and shortening time-to-market.

But, before diving deeper it is worth analyzing AIaaS stack which can be divided into three layers of capabilities (see Figure 2):

• AI Software Services that are ready-to-use AI applications and building blocks.
• AI Developer Services are tools for assisting developers in implementing code to bring out AI capabilities.
• AI Infrastructure Services that comprise raw computational power for building and training AI algorithms, and network and storage capacities to store and share data.

Another possible classification is the taxonomy summarized in Table 1. These categories reflect the perspective of users’ (decreasing) level of involvement, required technical expertise and control over the service and the underlying ML model(s), as well as the increasingly specialized types of problems they address. These categories are not mutually exclusive in that some services may have characteristics that overlap the service types; however, identifying the service types helps indicate and highlight certain properties and characteristics that can have fairness implications in practice.

At this point, we have enough context to understand the Human-Led AI Services approach (see Figure 3). Simply put, it focused on providing AI Services through a multilayered model, in the same sense that API-Led states for API connectivity, with each layer exposing its services through APIs.

At the lower level, we found System AI Services. They result from exposing ML Services in the form of APIs and provide the ability to work at very low level with a concrete platform or solution in the same way System APIs are used to decouple the backend. Also, at this level, we find the tools to work with generic models.

Process AI Services are oriented toward concrete business domains, features, and/or capabilities. Built on top of System AI Services they can provide a layer of modularity and specialization for example to build customized AI Services for concrete knowledge area that allows a fine-tuning of the underlaying ML platform.

On top, the Human Layer exists. It’s the layer where creativity resides. It allows combining different AI Services to foster innovation and provide innovative services by building them by combining Process AI Services.

Conclusions

As can be seen, when applying API-Led Connectivity to provide AI services, the benefits are manifold. Businesses can leverage AI functionalities offered by AIaaS platforms through APIs, gaining access to state-of-the-art AI models and algorithms. The modular approach provided by Human-Led AI Services allows for tailor-made solutions, ensuring that organizations only pay for the AI services they utilize, promoting cost-efficiency. API-Led Connectivity's standardized communication facilitates seamless integration, fostering collaboration and compatibility. Additionally, AI capabilities can be seamlessly embedded into applications, enriching user engagement.

References

Lins, S., Pandl, K.D., Teigeler, H.?et al.?Artificial Intelligence as a Service.?Bus Inf Syst Eng?63, 441–456 (2021). https://doi.org/10.1007/s12599-021-00708-w

Kornel Lewicki, Michelle Seng Ah Lee, Jennifer Cobbe, and Jatinder Singh. 2023. Out of Context: Investigating the Bias and Fairness Concerns of “Artificial Intelligence as a Service” In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (CHI '23). Association for Computing Machinery, New York, NY, USA, Article 135, 1–17. https://doi.org/10.1145/3544548.3581463

?